Dataset for paper "Illustrating Extreme Negative Linear Compressibility in Thermosalient Molecular Crystals"

gen2_9.model is a deployed MACE model. gen2_train.extxyz is the dataset containing 1,805 structures (three crystal forms) of N-2-Propylidene-4-hydroxybenzohydrazide, mostly unit cells but also some 2x2x1 supercells. These were created using molecular dynamics (MD) simulations under different temperatures and pressures. The dataset was used to train the model and is based on DFT calculations performed using the Vienna Ab-Initio Simulation Package, VASP 6.3.0, employing the r2SCAN exchange-correlation functional. (Note: Energy, forces, and stress have a prefix REF required by MACE for training.) Similarly, gen2_test.extxyz is the dataset containing 200 structures (under the same conditions and DFT settings as the training set) used to test the models after training. train.sh is the training script with all the hyperparameters used to create the model. For the Quasi-Harmonic Approximation (QHA), to run or reproduce the Gibbs free energy calculations as described in the paper: 1.cif is the XRD structure file of form 1. P_minimisation.py is the script used to create the structures for the energy-volume curve for the QHA calculations. freeE.py calculates the thermal properties in the Harmonic Approximation (HA) of the above-generated structures. pressure_structures_QHA.zip contains all the structures at different pressures calculated using QHA. The structures are numbered in order of increasing pressure and the pressure is the last number (in front of GPA) but with the decimal point after the first number (example: 1_f1_p005GPa.cif -> 0.05 GPa, 9_f1_tilt_p11GPa.cif -> 1.1 GPa ) The version of all necessary packages is as follows: - ase 3.24.0 - mace-torch 0.3.10 - python 3.13.2 - phonopy 2.37.0 - torch-dftd 0.5.1gen2_9.model 是一个已部署的MACE模型。 gen2_train.extxyz 是包含1805个N-2-亚丙基-4-羟基苯甲酰肼（N-2-Propylidene-4-hydroxybenzohydrazide）结构的数据集，涵盖三种晶型，其中绝大多数为原胞，同时包含少量2×2×1超胞。该数据集通过分子动力学（Molecular Dynamics，MD）模拟生成，模拟过程设置了不同的温度与压强条件。本数据集用于训练前述MACE模型，其数据基于维也纳从头算模拟包（Vienna Ab-Initio Simulation Package，VASP）6.3.0版本完成的密度泛函理论（Density Functional Theory，DFT）计算，计算时采用了r2SCAN交换关联泛函。（注：为满足MACE模型的训练要求，能量、受力与应力数据需添加REF前缀。） 与之类似，gen2_test.extxyz 是包含200个结构的测试数据集，其结构生成条件与DFT计算设置均与训练集保持一致，用于训练完成后对模型进行性能测试。 train.sh 是用于构建该模型的训练脚本，其中包含了所有训练超参数设置。 若需复现论文中所述的准谐近似（Quasi-Harmonic Approximation，QHA）吉布斯自由能计算流程，请执行以下内容： 1.cif 为晶型1的X射线衍射（X-ray Diffraction，XRD）结构文件。 P_minimisation.py 是用于生成准谐近似计算所需能量-体积曲线对应结构的脚本。 freeE.py 可基于上述生成的结构，在简谐近似（Harmonic Approximation，HA）框架下计算体系的热物性参数。 pressure_structures_QHA.zip 压缩包包含了所有通过准谐近似计算得到的不同压强下的结构文件。所有结构按压强递增顺序编号，文件名中GPA前的最后一组数字即为对应压强值，格式需将该数字的首位后添加小数点（示例：1_f1_p005GPa.cif 对应0.05 GPa，9_f1_tilt_p11GPa.cif 对应1.1 GPa） 本次实验所需全部软件包的版本信息如下： - ase 3.24.0 - mace-torch 0.3.10 - Python 3.13.2 - phonopy 2.37.0 - torch-dftd 0.5.1